This paper proposes a method to improve the quality of depth maps transmitted in multiple descriptions through multipath error prone networks. Whenever a single description is lost, the remaining ones are still able to provide a coarsely decoded version of the depth map. While in multiple description video, such coarse decoding is still acceptable for display, in the case of depth maps the additional decoding distortion propagates through the corresponding view synthesis. The proposed method is capable of enhancing low quality depth maps decoded from one single description, based on geometric information available in coarsely decoded slices, combined with higher quality depth values in adjacent slices decoded from both descriptions. In comparison with existing MDC decoders, the enhancement method achieves quality gains in synthesised views up to 1.69dB for packet loss rates of 10%.